Resources for Heinrich et al., 2021
The raw imaging data and ground truth data used for machine learning model training can be found in this s3 bucket:
s3://janelia-cosem-publications/heinrich-2021a
This directory contains array data stored hierarchically in the n5
format.
- FIB-SEM data used for generating ground truth and predictions is stored as n5 datasets, and can be found at
{dataset_name}/{dataset_name}.n5/volumes/raw
- Human-generated annotations ("crops") are stored as a collection of n5 groups in
{dataset_name}/{dataset_name}.n5/volumes/labels/0003/{crop_name}
. The pixel data for the crop is stored at{dataset_name}/{dataset_name}.n5/volumes/labels/0003/{crop_name}/labels/all
- Human-generated skeletons of microtubules used for validation of the microtubule refinements are available in
nml
format in{dataset_name}/skeletons/mt/block_{blockno}.nml
Warning: Be advised that copying large amounts of imaging data from cloud storage may use a lot of local storage space and take a long time to complete!
(AWS cli
) Download everything to local/path
.
aws s3 cp janelia-cosem-publications/heinrich-2021a/ local/path --recursive
(AWS cli
) Download just data for the jrc_hela-2
dataset to local/path/jrc_hela-2.n5
:
aws s3 cp janelia-cosem-publications/heinrich-2021a/jrc_hela-2/jrc_hela-2.n5 local/path/jrc_hela-2.n5 --recursive
- Training setups: scripts used for training ML networks on FIB-SEM data
- Training, Inference and Evaluation: codebase for training, applying and evaluating ML networks
- Manual evaluation of networks: Fiji plugin for manual evaluation of ML network predictions
- Segmentation and analysis: code and scripts used to segment predictions and analyze results; a list of all refinements used to get the final segmentations can be found here
- Watershed and agglomeration: scripts used to perform watershed segmentation and agglomeration on select organelles
- CLEM: scripts and tools used to co-register light and electron microscopic images.
- Microtubule Refinement: scripts and package used for microtubule refinement.